Document clustering as an unsupervised approach extensively used to navigate, filter, summarize and manage large collection of document repositories like the World Wide Web (WWW). Recently, focuses in this domain shifted from traditional vector based document similarity for clustering to suffix tree based document similarity, as it offers more semantic representation of the text present in the document. In this paper, we compare and contrast two recently introduced approaches to document clustering based on suffix tree data model. The first is an Efficient Phrase based document clustering, which extracts phrases from documents to form compact document representation and uses a similarity measure based on common suffix tree to cluster the documents. The second approach is a frequent word/word meaning sequence based document clustering, it similarly extracts the common word sequence from the document and uses the common sequence/ common word meaning sequence to perform the compact representation, and finally, it uses document clustering approach to cluster the compact documents. These algorithms are using agglomerative hierarchical document clustering to perform the actual clustering step, the difference in these approaches are mainly based on extraction of phrases, model representation as a compact document, and the similarity measures used for clustering. This paper investigates the computational aspect of the two algorithms, and the quality of results they produced.

Document clustering is usually performed as an unsupervised task. It attempts to separate different groups of documents (clusters) from a document collection based on implicitly identifying the common patterns present in these documents. A semi-supervised approach to this problem recently reported promising results. In semi-supervised approach, an explicit background knowledge (for example: Must-link or Cannot-link information for a pair of documents) is used in the form of constraints to drive the clustering process in the right direction. In this paper, a semi-supervised approach to document clustering is proposed. There are three main contributions through this paper (i) a document is transformed primarily into a graph representation based on Graph-of-Word approach. From this graph, a word sequences of size=3 is extracted. This sequence is used as a feature for the semi-supervised clustering. (ii) A similarity function based on commonword sequences is proposed, and (iii) the constrained based algorithm is designed to perform the actual cluster process through active learning. The proposed algorithm is implemented and extensively tested on three standard text mining datasets. The method clearly outperforms the recently proposed algorithms for document clustering in term of standard evaluation measures for document clustering task.

Document clustering is an unsupervised machine learning technique that organizes a large collection of documents into smaller, topic homogenous, meaningful sub-collections (clusters). Traditional document clustering approaches use extracted features like: word (term), phrases, sequences and topics from the documents as descriptors for clustering process. These features do not consider the relationship among different words that are used to convey the contextual information within the document. Recently, Graph-of-Word approach is introduced in information research; this approach addresses the problem of independence assumption by building a graph of word from the words that appeared in a document. Hence, the relationships among words are captured in the representation. It is an un[1]weighted directed graph whose vertices represent unique terms and whose edges represent co-occurrences between the terms. The representation is simplified by using a sliding window of size = 3 with the text of the document. This paper uses a sequence based-representation of document that is extracted from graph[1]of-word of the document. A similarity measure is defined over the common sequences between two documents. The proposed approach is implemented and tested on standard text mining datasets. A series of experiments reveal that the proposed approach outperforms the traditional approaches on clustering measures like: Purity, Entropy and F-Score.